Processing a subset of a feature set to determine whether to process a query request on a local artificial intelligence engine or a remote artificial intelligence engine

The present invention relates to a computer program product, system, and method for processing a subset of a feature set to determine whether to process a query request on a local artificial intelligence engine or a remote artificial intelligence engine.

A network system may maintain multiple edge servers capable of providing local Artificial Intelligence (AI) engines to process queries from client devices in close proximity to the edge servers, as opposed to sending the query to an AI engine on a remote server, which is not located close to where the user devices are. The edge AI engine may provide results with lower latency than sending the queries to a central server.

Provided are a computer program product, system, and method for processing a subset of a feature set to determine whether to process a query request on a local artificial intelligence engine or a remote artificial intelligence engine. A determination is made whether to process the query at a first machine learning model or a second machine learning model, wherein the second machine learning model comprises a larger and more complex model than the first machine learning model. One of the following is performed: 1) in response to the determining to process the query at the first machine learning model, forwarding non-sensitive input data of the query to the first machine learning model to produce a first query result to return to an initiator of the query; and 2) in response to determining to process the query at the second machine learning model, forwarding sensitive input data and the non-sensitive input data to an encryption engine to encrypt to send to the second machine learning model; and receiving an encrypted second query result from the second machine learning model to decrypt to produce a second query result to return to the initiator of the query.

Described embodiments provide improvements to computer technology to determine whether to process a query in a local inference or AI engine or send the query to a remote inference server to be processed by a larger machine learning model to provide better accuracy. Described embodiment perform AI inference under a privacy preserving technology that utilizes a privacy preserving remote AI engine. In described embodiments, sensitive and non-sensitive information may be segregated to allow privacy preservation of sensitive data. Processing at the remote inference server increases latency because sensitive data has to be encrypted before being transmitted. In order to trade off accuracy, performance and power consumption of the overall system, described embodiments provide a query scheduler comprising a machine learning model that analyzes a smaller subset of a feature set of a query, such as non-sensitive data, to determine whether a local AI engine is sufficient to process the query or whether greater accuracy and complexity is involved to prefer selecting to encrypt and transmit to a remote inference engine with a larger, more complex AI engine to process. In further embodiments, to use the more advanced models in a remote inference engine, the data will be encrypted and sent to the remote inference server for processing.

The term sensitive data as used herein refers to personal information, personal identifiable information, private information, personal financial information, and other information that comprises any representation of information that permits the identity of an individual to whom the information applies to be reasonably inferred by either direct or indirect means. Examples of personally identifiable information (PII) include: Social security number (SSN), passport number, driver's license number, taxpayer identification number, patient identification number, financial account or credit card number, etc. The term non-sensitive data refers to data that is not personally identifiable information that could identify a particular person, and comprises information not providing sensitive information on a person.

In certain implementations, an inference engine may be located at an edge server that may be in an on-premise location of an entity that is not privy to see certain sensitive information of their clients' data while performing analytics with the said data. In such a situation, a privacy-preserving technology, such as homomorphic encryption, may be used that allows computation on encrypted data without revealing the actual data. Due to the aforementioned latency constraints and because such privacy-preserving technologies typically require high computational resources, such an entity may offload the computation to a cloud server that performs the analytics under a privacy-preserving technology.

1 FIG. 102 200 104 200 106 107 108 110 109 110 illustrates an embodiment of a source server, such as a trusted initial processing center for a query for an inference engine, generating a query requestto send to a local inference server. The query requestmay be processed at a local artificial intelligence (“AI”) engine, such as a local machine learning model, to produce query resultsor forwarded to a remote inference serverto process the query at an AI under homomorphic encryption (“HE”) engineto produce query results. The AI enginemay comprise a neural network that performs the neural network operations using homomorphic encryption to allow the processing directly on the homomorphic encrypted data.

112 114 114 116 114 112 116 118 120 112 122 118 114 124 112 200 202 124 120 204 114 2 FIG. The source server includes a feature extraction engineto receive a user queryincluding a feature set comprising the input data for the queryand user privacy preferences. The querymay comprise a question, such as a transcribed telephone conversation with a customer or query data in other formats. The feature extraction engineuses the privacy preferencesto distinguish between sensitive data, i.e., features, and non-sensitive data, e.g., features. Sensitive data may comprise personal information, such as customer contact and billing information and non-sensitive information may comprise general information, such as make and model of a product. In further embodiments, the feature extraction enginemay categorize the data into other categories than sensitive or non-sensitive. A symmetric encryption engine, implementing a symmetric encryption algorithm, e.g., AES (Advanced Encryption Standard), DES (Data Encryption Standard), Blowfish, etc., encrypts the sensitive data, which may comprise highly personal data for the user that would be helpful in obtaining a desired query result for the query, into encrypted sensitive data. The feature extraction enginebuilds a query request, shown in, having a query identifier (ID), the encrypted sensitive data, the non-sensitive data, which is cleartext, and a criticality settingprovided by a user initiating the queryindicating a critical element in supplying the query results, such as accuracy, latency, or both accuracy and latency.

110 106 106 110 124 120 108 104 110 106 In certain embodiments, the AI under HE enginemay comprise a larger and more complex machine learning model than the local AI engine. For instance, the local AI enginemay process only the non-sensitive data, i.e., features, whereas the remote AI enginemay process the sensitive data, the non-sensitive data, and additional encrypted sensitive information maintained in a large database in the remote inference serverto enrich the query data with sensitive information that might improve the accuracy of the model. This additional sensitive input to enrich the query data may not be available at the local inference server. In this way, the remote AI enginereceives as input a larger feature set than the local AI engineto produce more robust and accurate results.

104 126 120 200 200 106 200 110 108 106 120 The local inference serverincludes a query scheduler, comprising a machine learning model, trained to process the non-sensitive input datato generate a complexity score indicating an extent of complexity needed to process the query request. In certain embodiments, if the complexity score is below a low threshold, then this indicates the query requestis of relatively low complexity and may be processed at the local AI engine. If the complexity score exceeds a high complexity score threshold, this this indicates the query requestrequires relatively more complex processing and should be processed at the remote AI engine, which receives as input a larger feature set including sensitive and non-sensitive data and optionally encrypted sensitive information maintained in a large database in the remote inference server, whereas the local AI enginemay only process non-sensitive input data. If the output complexity score is between the low and high complexity thresholds, then additional information on user preference for low latency and/or accuracy may be considered to determine where the query is processed.

127 126 104 108 108 108 106 106 108 108 In additional embodiments, the local inference server may include a complexity score adjusterto receive a base complexity score from the query schedulerand adjust the complexity score from metrics related to the latency/energy/power of performing the task at the local inference server, latency in network transmission to the remote inference server, current level of congestion in the network, and latency/energy/power of performing the task at the remote inference server, to produce an adjusted complexity score. For instance, if the latency at the remove inference serveror in network transmission is considered high, then the base complexity score may be adjusted lower to increase the likelihood of processing at the local inference server. Alternatively, if latency at the local inference serveris relatively high as compared to processing at the remote inference server, then the base complexity score may be adjusted upward to increase likelihood of transmitting to the remote inference serverfor processing.

127 104 108 126 127 127 The complexity score adjustermay comprise a small machine learning model, such as a simple decision tree or multilayer perceptron, to obtain a score based on input such as number of words in the query, maximum sentence length, language, etc., as well as latency conditions for processing at the local inference serverand remote inference server. In described embodiments, the “complexity score” may refer to the complexity score from the query schedulerwithout adjustments by a complexity score adjusteror may refer to the adjusted complexity score produced by the complexity score adjuster.

104 128 130 120 124 132 300 108 134 300 302 304 306 308 126 110 108 300 3 FIG. The local inference serverincludes a homomorphic encryption (HE) enginehaving an HE encoderto convert the input data,, which may comprise integers, real or complex numbers, into polynomials, and an HE encryption engineto perform homomorphic encryption on the encoded polynomials to produce HE encrypted query inputof input data/features to forward to the remote inference serverover a network, such as the Internet.provides an example of the HE encrypted query inputas including a query ID, HE encrypted symmetric encrypted sensitive data, HE encrypted non-sensitive data, and a selected AI under HE engineif the query schedulerselects one of a plurality of remote AI under HE enginesat the remote inference serverto process the HE encrypted query input.

128 136 138 110 140 109 102 The HE enginefurther includes an HE decryption engineto process HE encrypted results, produced by the AI under HE engine, to produce unencrypted query results represented as polynomials. An HE decoderdecodes the unencrypted polynomials representing the query results into clear text query results, to return to the source serverand query.

108 142 110 144 304 302 110 138 The remote inference serverincludes an HE evaluation enginethat includes the AI under HE engineand a symmetric decryption engine under homomorphic encryption (HE)to perform symmetric decryption under HE of the symmetric encrypted sensitive datato produce HE encrypted sensitive data, without symmetric encryption, represented by encrypted polynomials. The HE encrypted sensitive data is forwarded along with the HE encrypted non-sensitive datato the AI under HE engineto produce the HE encrypted results.

104 400 400 402 120 126 107 106 109 110 406 146 400 126 109 110 406 107 106 406 i 4 FIG. The local inference server, or another server, may gather training records, including a training record instance, as described with respect to, having a query ID, the non-sensitive input dataprocessed by the query schedulerto generate the complexity score, the query resultsfrom the local AI engine, query resultsfrom the remote AI under HE engine, and a ground truth resultfor the query. A query scheduler training enginemay use the training recordsto train the query scheduler, such as using backward propagation or other cost function optimization techniques, to output a high complexity score if the query resultfrom the remote AI engineis the ground truth resultor to output a low complexity score if the query resultfrom the local AI engineis the ground truth result.

104 108 104 108 104 108 108 In described embodiments, the local inference serverencrypts input data to send to the remote inference serverusing homomorphic encryption. In alternative embodiments, different encryption techniques, such as Pallier encryption or garbled circuits, may be used at the local inference serverto encrypt input data to send to the remote inference server. In certain embodiments, neither the local inference servernor the remote inference servermay not be able to access the sensitive data in the clear. Additionally, remote inference servermay not be able to access the non-sensitive data in the clear.

104 108 108 104 The localand the remoteinference servers may comprise enterprise level servers. In certain embodiment, the remote inference servermay comprise a more powerful server with greater computational resources than the local inference server.

106 110 112 122 126 128 130 132 136 140 142 144 146 Generally, program modules, such as the program components,,,,,,,,,,,,, among others, may comprise routines, programs, objects, components, logic, data structures, and so on that perform particular tasks or implement particular abstract data types.

106 110 112 122 126 128 130 132 136 140 142 144 146 The program components programs,,,,,,,,,,,,, among others, may comprise program code loaded into memory and executed by a processor. Alternatively, some or all of the program logic of these components may be implemented in hardware devices, such as in Application Specific Integrated Circuits (ASICs), Field Programmable Gate Arrays (FPGAs), Graphics Processing Units (GPUs), and/or AI hardware accelerator.

106 110 112 122 126 128 130 132 136 140 142 144 146 The functions described as performed by the program components,,,,,,,,,,,,, among others, may be implemented as program code or hardware logic in fewer program modules than shown or implemented throughout a greater number of program modules than shown.

102 104 108 The servers,,may comprise server class computing devices, or other suitable computing devices.

1 FIG. 106 110 112 122 126 128 130 132 136 140 142 144 146 102 104 108 In, arrows are shown between components,,,,,,,,,,,,, among others, in the servers,,. These arrows represent information flow to and from the program components.

102 104 108 104 134 104 102 In one embodiment, the source serverand the local inference serverare in close proximity in the same or different premises, such as on a Local Area Network (LAN). The remote servermay comprise a cloud server accessible to the local inference serverover a network, such as the Internet. In certain implementations, the local inference servermay comprise an edge server providing AI query processing to one or more source serversin close proximity to reduce latency for certain query results.

106 110 126 106 110 126 106 110 126 106 110 126 Certain of the program components, such as,,, may use machine learning and deep learning algorithms, such as decision tree learning, association rule learning, neural network, inductive programming logic, support vector machines, Bayesian network, Recurrent Neural Networks (RNN), Feedforward Neural Networks, Convolutional Neural Networks (CNN), Deep Convolutional Neural Networks (DCNNs), Generative Adversarial Network (GAN), etc. For artificial neural network program implementations, the neural network may be trained using backward propagation to adjust weights and biases at nodes in a hidden layer to produce their output based on the received inputs. In backward propagation used to train a neural network machine learning module, biases at nodes in the hidden layer are adjusted accordingly to produce the output having specified confidence levels based on the input parameters. The machine learning models,,, may be trained to produce their output based on the inputs. Backward propagation may comprise an algorithm for supervised learning of artificial neural networks using gradient descent. Given an artificial neural network and an error function, the method may use gradient descent to find the parameters (coefficients) for the nodes in a neural network or function that minimizes a cost function measuring the difference or error between actual and predicted values for different parameters. The parameters are continually adjusted during gradient descent to minimize the error. Other techniques may be used to train the machine learning models,,to adjust the biases and weights to minimize error between ground truth results from the machine learning model,,and actual output.

106 110 126 106 110 126 In an alternative embodiment, the components,,, may be implemented not as a machine learning model but implemented using a rules-based system to determine the outputs from the inputs. The components,,, may further be implemented using an unsupervised machine learning module, or machine learning implemented in methods other than neural networks, such as multivariable linear regression models.

Components implemented as a machine learning model may be implemented in programs in memory or in a hardware accelerator or an inference engine.

5 FIG. 102 200 104 500 114 116 112 502 118 120 122 504 118 124 120 120 200 506 202 124 120 204 illustrates an embodiment of operations performed at the source serverto generate a query requestto send to the local inference server. Upon receiving (at block) a queryfrom a user with specified user privacy preferences, the feature extraction engineextracts (at block) sensitiveand non-sensitivedata, i.e., features. The symmetric encryption engineencrypts (at block) the sensitive data, using a symmetric encryption algorithm, to produce symmetric encrypted sensitive data. This leaves the non-sensitive datacleartext/unencrypted and the sensitive dataencrypted. A query requestis generated (at block) including a query ID, symmetric encrypted sensitive data, non-sensitive data, and a criticality settingindicating whether accuracy and/or low latency are requested in processing the query request.

4 FIG. 118 102 With the embodiment of, the sensitive datais encrypted to prevent access while transmitted from the source serverto protect personal information and comply with any legal requirements with respect to handing personal information.

6 FIG. 126 200 106 110 600 200 126 602 120 200 604 200 126 606 120 106 107 200 604 608 126 610 120 124 130 120 124 illustrates an embodiment of operations performed by the query schedulerto determine whether to send the query requestto the local AI engineor the remote AI engine. Upon receiving (at block) the query request, the query schedulermachine learning model processes (at block) the non-sensitive input datato produce a complexity score indicating an extent to which the query requestrequires low or high complexity to process. If (at block) the complexity score is less than a low threshold, i.e., a querynot requiring more complex processing, then the query schedulerforwards (at block) the non-sensitive input datato the local AI engineto produce a query resultto return to the query request. If (at block) the complexity score is greater than the low threshold and if (at block) the complexity score is higher than a high complexity score threshold, which is higher than the low complexity threshold, then the query schedulerforwards (at block) the non-sensitive input dataand the symmetric encrypted sensitive datato a homomorphic encoderto produce a polynomial representing the input data/features,.

604 608 127 126 106 108 In further embodiments, the complexity score considered at blocksandmay comprise an adjusted complexity score from the complexity score adjusterthat adjusts a base complexity score from the query schedulerto adjust for latency/energy/power conditions at the local inference severand the remote inference serverto produce an adjusted complexity score.

120 124 612 132 306 304 300 614 304 306 308 126 108 110 The polynomials representing the non-sensitive input dataand the symmetric encrypted sensitive dataare forwarded (at block) to a homomorphic encryption (HE) engineto generate encrypted polynomials representing the HE encrypted non-sensitive dataand the HE encrypted symmetrically encrypted sensitive data. A queryis generated (at block) including the HE encrypted non-sensitive data, the HE encrypted symmetrically encrypted sensitive data, and an optional selected neural network model for the AI under HE engineif outputted by the query scheduler, and sent to the remote inference serverfor processing by the AI under HE engine.

126 108 110 108 126 In alternative embodiments, the query schedulermay further predict which of multiple remote neural network models should be used to process the query on the remote inference server. For instance, if there are four neural network modelson the remote inference server, the query schedulermay output which of the several neural network models to use for the remote query. The query scheduler may output one of multiple advanced analytic processing categories which corresponds to a complexity of the remote neural network model, relative to complexity of other of the remote models. These may trade-off, for instance, high accuracy of inference for a longer latency.

7 FIG. 142 108 142 700 300 306 304 144 702 304 306 704 110 138 138 706 104 illustrates an embodiment of operations performed by the HE evaluation enginein the remote inference server. Upon the HE evaluation enginereceiving (at block) the HE encrypted querywith the HE encrypted non-sensitive dataand HE encrypted-symmetrically encrypted sensitive data, the symmetric decrypt engine under HEprocesses (at block) the HE encrypted symmetrically encrypted sensitive datato produce HE encrypted sensitive data. The HE encrypted non-sensitive dataand the HE encrypted sensitive data are inputted (at block) to the AI under HE encryption engineto produce an HE encrypted query result. The HE encrypted query resultis returned (at block) to the local inference server.

118 122 118 122 144 302 110 102 104 108 102 104 In described embodiments, the sensitive input datais subject to encryption by the symmetric encryption engine. In an alternative embodiment, the sensitive input datamay not be encrypted with a symmetric encryption algorithm. In such alternative embodiments, the HE encrypted sensitive data is not subject to symmetric encryption and may be only subject to HE encryption. In such alternative embodiment, there is no need for the symmetric encryption engineand the symmetric decrypt engine under HE, and the HE encrypted sensitive data along with the HE encrypted non-sensitive datamay be forwarded to the AI under HE engineto operate on the HE encrypted data. In embodiments where there is no symmetric encryption, the HE encode/encrypt may occur at a trusted source serverbefore transmission in order to prevent the local inference serverand remote inference serverfrom accessing the sensitive data in the clear. In this embodiment, the HE encode/encrypt for the non-sensitive data may happen either at the trusted source serveror the local inference server.

108 110 104 108 104 104 108 110 110 108 In further embodiments, the remote inference servermay have access to an encrypted big data database that can be used to enrich the query data and provide additional sensitive information that might improve the accuracy of the AI engine. This additional information may not be available at the local inference server. Such an encrypted database on the remote inference servermay be populated with encrypted information from the local inference server, in addition to information from local inference serversfrom other entities, such as on different edge servers. This encrypted database on the remote inference servermay be used to either (i) fine-tune the remote AI under HE engineto improve the accuracy of neural networks running on the engine, or (ii) appended additional features to the on-line query for each inference task that arrives at the remote inference server.

8 FIG. 128 138 128 800 138 136 802 109 140 804 114 illustrates an embodiment of operations performed by the HE engineto decrypt and the query resultto return in cleartext to the initiator of the query. Upon the HE enginereceiving (at block) the HE encrypted result, the HE decryption engineproduces (at block) unencrypted polynomial(s) representing the query result. The HE decoderproduces (at block) the cleartext query result, such as an integer vector, to return to the queryinitiator.

9 FIG. 6 FIG. 3 FIG. 126 106 110 200 616 900 126 127 902 204 200 126 904 606 610 612 614 107 109 106 110 126 906 106 110 200 126 908 106 110 illustrates an embodiment of operations performed by the query schedulerto determine to have the local AI engineor remote AI under HE engineprocess the query requestwhen called at blockinin the event the complexity score is between the low and the high complexity score threshold. Upon initiating processing (at block) when the complexity score, which may comprise a base complexity score from the query scheduleror adjusted complexity score from the complexity score adjuster, is between the low and the high thresholds, if (at block) the criticalityin the query requestindicates low latency processing, then the query schedulerperforms (at block) operations at blocks,,into obtain query results,from both the local AI engineand from remote AI under HE engine. The query schedulerreturns (at block) a response from one of the local AI engineand the remote AI under HE enginethat is first to return a response to the query request. The query schedulersends (at block) stop signals to the AI engineor the AI under HE enginethat did not return a response to the query first to conserve energy and computational resources.

902 204 126 912 200 106 914 107 107 916 114 914 107 126 918 610 612 614 109 110 920 109 109 922 114 920 109 107 109 924 3 FIG. If (from the NO branch at block) criticalityis energy, or most energy efficient processing, then the query schedulerforwards (at block) the query requestto the local AI engineto process. If (at block) the local query resulthas a confidence level exceeding a confidence threshold, then the local query resultis returned (at block) to the query. If (at block) the local query resultdoes not exceed the confidence level threshold, then the query schedulerperforms (at block) operations at blocks,,into obtain a remote query resultfrom the remote AI under HE engine. If (at block) the remote query resulthas a confidence level exceeding a confidence threshold, then the remote query resultis returned (at block) to the query. If (at block) the remote query resultconfidence level also does not exceed the confidence threshold, then the localor remotequery threshold having a highest confidence level is returned (at block).

912 106 110 106 110 106 110 106 110 912 106 110 9 FIG. The energy efficient processing at blockinassumes the local query engineconsumes less energy than the AI under HE engine, so the local query engineis considered first for processing. In an embodiment where the AI under HE engineconsumes less energy than the local query engine, the AI under HE enginemay process the query first and the local query engineconsidered second if the confidence level of the result of the AI under HE engineis not sufficient. Alternatively, there may be first a check before blockof what is the more energy efficient engineor, and that more energy efficient engine may be selected first to process.

107 109 The confidence level of a local query resultor remote query resultmay be a score indicating the likelihood that the query result is correct, given that the ground truth information may not be available immediately at the time of processing the query. The confidence or confidence level of an inference may be computed in multiple ways, such as maximum probability (highest probability in the output vector of the outcome being close to 1 may indicate a better model), entropy (high entropy may indicate a better model), variance (higher variance from a mean value may indicate a better model), etc.

910 204 126 926 606 610 612 614 107 109 106 110 107 109 126 928 107 109 107 109 930 922 924 930 107 109 400 402 120 126 404 127 107 109 406 400 126 406 3 FIG. i i If (at block) criticalityindicates only accuracy, then the query schedulerperforms (at block) the operations at blocks,,into obtain query resultsandfrom both the local AI engineand from remote AI under HE engine, respectively. Upon receiving both query resultsand, the query schedulerdetermines (at block) the confidence of the query results,. The query result,having the highest determined confidence is returned (at block). For those situations at blocks,, andwhere both query results,were obtained, a training recordis created indicating the query; the non-sensitive input datathe query schedulerprocesses; the complexity score, or adjusted complexity score from the complexity score adjuster; the local query result; and the remote query result. The ground truth resultwill be received later. The gathered training recordsare used to train the query schedulermachine learning model once the ground truth resultis received for that query.

10 FIG. 146 126 400 146 104 107 109 400 400 400 406 107 106 126 1008 120 1010 1006 107 406 1012 109 126 1014 120 1016 1018 1004 400 i i i illustrates an embodiment of operations performed by the query scheduler training engineto train the query schedulerwith the training records. The training enginemay reside on the local inference serveror another location, such as a developer site. Upon receiving ground truth results for queries having query resultsandstored in training records, the ground truth results are saved in the training records. For each training recordhaving ground truth results, if the local query resultfrom the local AI engineis the ground truth result, then the query schedulermachine learning model is trained (at block) to process non-sensitive input datato output a lowest complexity score, e.g., 0. The low complexity score threshold is incremented (at block) by some amount, such as a trainable hyperparameter or by a fixed amount. If (at block) the local query resultis not the ground truth resultand if (at block) the remote query resultis the ground truth result, then the query schedulermachine learning model is trained (at block) to process non-sensitive input datato output a highest complexity score, e.g., 1. The high complexity score threshold is decremented (at block) by some amount. At block, control proceeds back to blockto process a next training recordhaving a ground truth result.

10 FIG. 126 120 107 109 126 106 110 106 110 With the embodiment of, the query scheduleris trained, based on the input non-sensitive datato output a value closest to the lowest complexity score when the local query resultis correct and is trained to output a value closest to the highest complexity score when the remote query resultis correct. This trains the query schedulerto produce more accurate selection of the local AI engineor remote AI under HE engineto use to process the query result. In described embodiments, the low complexity score threshold is incremented and high complexity threshold decremented to reduce the window between the low and the high thresholds as much as possible to achieve the goal of directing most queries to one of the local AI engineor the remote AI under HE engine, which requires less processing than the case where the complexity score is between the low and high thresholds. Further, in certain embodiments, an incoming query may be processed as if its complexity score was between high and low thresholds even when the complexity score is not between the high and low thresholds. This may be performed at fixed or trainable time interval parameters.

The present invention may be a system, a method, and/or a computer program product. The computer program product may include a computer-readable storage medium (or media) having computer-readable program instructions thereon for causing a processor to carry out aspects of the present invention.

Various aspects of the present disclosure are described by narrative text, flowcharts, block diagrams of computer systems and/or block diagrams of the machine logic included in computer program product (CPP) embodiments. With respect to any flowcharts, depending upon the technology involved, the operations can be performed in a different order than what is shown in a given flowchart. For example, again depending upon the technology involved, two operations shown in successive flowchart blocks may be performed in reverse order, as a single integrated step, concurrently, or in a manner at least partially overlapping in time.

A computer program product embodiment (“CPP embodiment” or “CPP”) is a term used in the present disclosure to describe any set of one, or more, storage media (also called “mediums”) collectively included in a set of one, or more, storage devices that collectively include machine readable code corresponding to instructions and/or data for performing computer operations specified in a given CPP claim. A “storage device” is any tangible device that can retain and store instructions for use by a computer processor. Without limitation, the computer-readable storage medium may be an electronic storage medium, a magnetic storage medium, an optical storage medium, an electromagnetic storage medium, a semiconductor storage medium, a mechanical storage medium, or any suitable combination of the foregoing. Some known types of storage devices that include these mediums include: diskette, hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or Flash memory), static random access memory (SRAM), compact disc read-only memory (CD-ROM), digital versatile disk (DVD), memory stick, floppy disk, mechanically encoded device (such as punch cards or pits/lands formed in a major surface of a disc) or any suitable combination of the foregoing. A computer-readable storage medium, as that term is used in the present disclosure, is not to be construed as storage in the form of transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide, light pulses passing through a fiber optic cable, electrical signals communicated through a wire, and/or other transmission media. As will be understood by those of skill in the art, data is typically moved at some occasional points in time during normal operations of a storage device, such as during access, de-fragmentation or garbage collection, but this does not render the storage device as transitory because the data is not transitory while it is stored.

11 FIG. 1100 1101 1145 126 106 128 1100 1101 1102 1103 1104 1105 1106 1101 1110 1120 1121 1111 1112 1113 1122 1145 1114 1123 1124 1125 1115 1104 1130 1105 1140 1141 1142 1143 1144 With respect to, computing environmentcontains an example of an environment for the execution of at least some of the computer code involved in performing the inventive methods, such as having a query scheduler select to process a query request in a local inference server or remote inference server. The computerincludes a hardware accelerator or graphical processing unit (GPU)having multiple hardware components, including a query scheduler, AI engine, and HE engineas described above. In addition, the computing environmentincludes, for example, computer, wide area network (WAN), end user device (EUD), remote server, public cloud, and private cloud. In this embodiment, computerincludes processor set(including processing circuitryand cache), communication fabric, volatile memory, persistent storage(including operating systemand hardware accelerator, as identified above), peripheral device set(including user interface (UI) device set, storage, and Internet of Things (IoT) sensor set), and network module. Remote serverincludes remote database. Public cloudincludes gateway, cloud orchestration module, host physical machine set, virtual machine set, and container set.

1101 1130 1100 1101 1101 1101 11 FIG. COMPUTERmay take the form of a desktop computer, laptop computer, tablet computer, smart phone, smart watch or other wearable computer, mainframe computer, quantum computer or any other form of computer or mobile device now known or to be developed in the future that is capable of running a program, accessing a network or querying a database, such as remote database. As is well understood in the art of computer technology, and depending upon the technology, performance of a computer-implemented method may be distributed among multiple computers and/or between multiple locations. On the other hand, in this presentation of computing environment, detailed discussion is focused on a single computer, specifically computer, to keep the presentation as simple as possible. Computermay be located in a cloud, even though it is not shown in a cloud in. On the other hand, computeris not required to be in a cloud except to any extent as may be affirmatively indicated.

1110 1120 1120 1121 1110 1110 PROCESSOR SETincludes one, or more, computer processors of any type now known or to be developed in the future. Processing circuitrymay be distributed over multiple packages, for example, multiple, coordinated integrated circuit chips. Processing circuitrymay implement multiple processor threads and/or multiple processor cores. Cacheis memory that is located in the processor chip package(s) and is typically used for data or code that should be available for rapid access by the threads or cores running on processor set. Cache memories are typically organized into multiple levels depending upon relative proximity to the processing circuitry. Alternatively, some, or all, of the cache for the processor set may be located “off chip.” In some computing environments, processor setmay be designed for working with qubits and performing quantum computing.

1101 1110 1101 1121 1110 1100 1145 Computer-readable program instructions are typically loaded onto computerto cause a series of operational steps to be performed by processor setof computerand thereby effect a computer-implemented method, such that the instructions thus executed will instantiate the methods specified in flowcharts and/or narrative descriptions of computer-implemented methods included in this document (collectively referred to as “the inventive methods”). These computer-readable program instructions are stored in various types of computer-readable storage media, such as cacheand the other storage media discussed below. The program instructions, and associated data, are accessed by processor setto control and direct performance of the inventive methods. In computing environment, at least some of the instructions for performing the inventive methods may be implemented in hardware accelerator.

1111 1101 COMMUNICATION FABRICis the signal conduction path that allows the various components of computerto communicate with each other. Typically, this fabric is made of switches and electrically conductive paths, such as the switches and electrically conductive paths that make up buses, bridges, physical input/output ports and the like. Other types of signal communication paths may be used, such as fiber optic communication paths and/or wireless communication paths.

1112 1112 1101 1112 1101 1101 VOLATILE MEMORYis any type of volatile memory now known or to be developed in the future. Examples include dynamic type random access memory (RAM) or static type RAM. Typically, volatile memoryis characterized by random access, but this is not required unless affirmatively indicated. In computer, the volatile memoryis located in a single package and is internal to computer, but, alternatively or additionally, the volatile memory may be distributed over multiple packages and/or located externally with respect to computer.

1113 1101 1113 1113 1122 1145 PERSISTENT STORAGEis any form of non-volatile storage for computers that is now known or to be developed in the future. The non-volatility of this storage means that the stored data is maintained regardless of whether power is being supplied to computerand/or directly to persistent storage. Persistent storagemay be a read only memory (ROM), but typically at least a portion of the persistent storage allows writing of data, deletion of data and re-writing of data. Some familiar forms of persistent storage include magnetic disks and solid state storage devices. Operating systemmay take several forms, such as various known proprietary operating systems or open source Portable Operating System Interface-type operating systems that employ a kernel. The hardware logic code included in the hardware acceleratortypically includes at least some of the hardware logic in performing the inventive methods.

1114 1101 1101 1123 1124 1124 1124 1101 1101 1125 PERIPHERAL DEVICE SETincludes the set of peripheral devices of computer. Data communication connections between the peripheral devices and the other components of computermay be implemented in various ways, such as Bluetooth connections, Near-Field Communication (NFC) connections, connections made by cables (such as universal serial bus (USB) type cables), insertion-type connections (for example, secure digital (SD) card), connections made through local area communication networks and even connections made through wide area networks such as the internet. In various embodiments, UI device setmay include components such as a display screen, speaker, microphone, wearable devices (such as goggles and smart watches), keyboard, mouse, printer, touchpad, game controllers, and haptic devices. Storageis external storage, such as an external hard drive, or insertable storage, such as an SD card. Storagemay be persistent and/or volatile. In some embodiments, storagemay take the form of a quantum computing storage device for storing data in the form of qubits. In embodiments where computeris required to have a large amount of storage (for example, where computerlocally stores and manages a large database) then this storage may be provided by peripheral storage devices designed for storing very large amounts of data, such as a storage area network (SAN) that is shared by multiple, geographically distributed computers. IoT sensor setis made up of sensors that can be used in Internet of Things applications. For example, one sensor may be a thermometer and another sensor may be a motion detector.

1115 1101 1102 1115 1115 1115 1101 1115 NETWORK MODULEis the collection of computer software, hardware, and firmware that allows computerto communicate with other computers through WAN. Network modulemay include hardware, such as modems or Wi-Fi signal transceivers, software for packetizing and/or de-packetizing data for communication network transmission, and/or web browser software for communicating data over the internet. In some embodiments, network control functions and network forwarding functions of network moduleare performed on the same physical hardware device. In other embodiments (for example, embodiments that utilize software-defined networking (SDN)), the control functions and the forwarding functions of network moduleare performed on physically separate devices, such that the control functions manage several different network hardware devices. Computer-readable program instructions for performing the inventive methods can typically be downloaded to computerfrom an external computer or external storage device through a network adapter card or network interface included in network module.

1102 1102 WANis any wide area network (for example, the internet) capable of communicating computer data over non-local distances by any technology for communicating computer data, now known or to be developed in the future. In some embodiments, the WANmay be replaced and/or supplemented by local area networks (LANs) designed to communicate data between devices located in a local area, such as a Wi-Fi network. The WAN and/or LANs typically include computer hardware such as copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and edge servers.

1103 1101 1101 1103 1101 1101 1115 1101 1102 1103 1103 1103 1103 102 END USER DEVICE (EUD)is any computer system that is used and controlled by an end user (for example, a customer of an enterprise that operates computer), and may take any of the forms discussed above in connection with computer. EUDtypically receives helpful and useful data from the operations of computer. For example, in a hypothetical case where computeris designed to provide a recommendation to an end user, this recommendation would typically be communicated from network moduleof computerthrough WANto EUD. In this way, EUDcan display, or otherwise present, the recommendation to an end user. In some embodiments, EUDmay be a client device, such as thin client, heavy client, mainframe computer, desktop computer and so on. In described embodiments, the EUDmay comprise the source serverdescribed above.

1104 1101 1104 1101 1104 1101 1101 1101 1130 1104 1104 108 REMOTE SERVERis any computer system that serves at least some data and/or functionality to computer. Remote servermay be controlled and used by the same entity that operates computer. Remote serverrepresents the machine(s) that collect and store helpful and useful data for use by other computers, such as computer. For example, in a hypothetical case where computeris designed and programmed to provide a recommendation based on historical data, then this historical data may be provided to computerfrom remote databaseof remote server. In described embodiments, the remote servermay comprise the remote inference serverand components therein as described above.

1105 1105 1141 1105 1142 1105 1143 1144 1141 1140 1105 1102 PUBLIC CLOUDis any computer system available for use by multiple entities that provides on-demand availability of computer system resources and/or other computer capabilities, especially data storage (cloud storage) and computing power, without direct active management by the user. Cloud computing typically leverages sharing of resources to achieve coherence and economies of scale. The direct and active management of the computing resources of public cloudis performed by the computer hardware and/or software of cloud orchestration module. The computing resources provided by public cloudare typically implemented by virtual computing environments that run on various computers making up the computers of host physical machine set, which is the universe of physical computers in and/or available to public cloud. The virtual computing environments (VCEs) typically take the form of virtual machines from virtual machine setand/or containers from container set. It is understood that these VCEs may be stored as images and may be transferred among and between the various physical machine hosts, either as images or after instantiation of the VCE. Cloud orchestration modulemanages the transfer and storage of images, deploys new instantiations of VCEs and manages active instantiations of VCE deployments. Gatewayis the collection of computer software, hardware, and firmware that allows public cloudto communicate through WAN.

Some further explanation of virtualized computing environments (VCEs) will now be provided. VCEs can be stored as “images.” A new active instance of the VCE can be instantiated from the image. Two familiar types of VCEs are virtual machines and containers. A container is a VCE that uses operating-system-level virtualization. This refers to an operating system feature in which the kernel allows the existence of multiple isolated user-space instances, called containers. These isolated user-space instances typically behave as real computers from the point of view of programs running in them. A computer program running on an ordinary operating system can utilize all resources of that computer, such as connected devices, files and folders, network shares, CPU power, and quantifiable hardware capabilities. However, programs running inside a container can only use the contents of the container and devices assigned to the container, a feature which is known as containerization.

1106 1105 1106 1102 1105 1106 PRIVATE CLOUDis similar to public cloud, except that the computing resources are only available for use by a single enterprise. While private cloudis depicted as being in communication with WAN, in other embodiments a private cloud may be disconnected from the internet entirely and only accessible through a local/private network. A hybrid cloud is a composition of multiple clouds of different types (for example, private, community or public cloud types), often respectively implemented by different vendors. Each of the multiple clouds remains a separate and discrete entity, but the larger hybrid cloud architecture is bound together by standardized or proprietary technology that enables orchestration, management, and/or data/application portability between the multiple constituent clouds. In this embodiment, public cloudand private cloudare both part of a larger hybrid cloud.

11 FIG. 1106 CLOUD COMPUTING SERVICES AND/OR MICROSERVICES (not separately shown in): private and public cloudsare programmed and configured to deliver cloud computing services and/or microservices (unless otherwise indicated, the word “microservices” shall be interpreted as inclusive of larger “services” regardless of size). Cloud services are infrastructure, platforms, or software that are typically hosted by third-party providers and made available to users through the internet. Cloud services facilitate the flow of user data from front-end clients (for example, user-side servers, tablets, desktops, laptops), through the internet, to the provider's systems, and back. In some embodiments, cloud services may be configured and orchestrated according to as “as a service” technology paradigm where something is being presented to an internal or external customer in the form of a cloud computing service. As-a-Service offerings typically provide endpoints with which various customers interface. These endpoints are typically based on a set of APIs. One category of as-a-service offering is Platform as a Service (PaaS), where a service provider provisions, instantiates, runs, and manages a modular bundle of code that customers can use to instantiate a computing platform and one or more applications, without the complexity of building and maintaining the infrastructure typically associated with these things. Another category is Software as a Service (SaaS) where software is centrally hosted and allocated on a subscription basis. SaaS is also known as on-demand software, web-based software, or web-hosted software. Four technological sub-fields involved in cloud services are: deployment, integration, on demand, and virtual private networks.

The letter designators, such as i, among others, are used to designate an instance of an element, i.e., a given element, or a variable number of instances of that element when used with the same or different elements.

The terms “an embodiment”, “embodiment”, “embodiments”, “the embodiment”, “the embodiments”, “one or more embodiments”, “some embodiments”, and “one embodiment” mean “one or more (but not all) embodiments of the present invention(s)” unless expressly specified otherwise.

The terms “including”, “comprising”, “having” and variations thereof mean “including but not limited to”, unless expressly specified otherwise.

The enumerated listing of items does not imply that any or all of the items are mutually exclusive, unless expressly specified otherwise.

The terms “a”, “an” and “the” mean “one or more”, unless expressly specified otherwise.

Devices that are in communication with each other need not be in continuous communication with each other, unless expressly specified otherwise. In addition, devices that are in communication with each other may communicate directly or indirectly through one or more intermediaries.

A description of an embodiment with several components in communication with each other does not imply that all such components are required. On the contrary a variety of optional components are described to illustrate the wide variety of possible embodiments of the present invention.

When a single device or article is described herein, it will be readily apparent that more than one device/article (whether or not they cooperate) may be used in place of a single device/article. Similarly, where more than one device or article is described herein (whether or not they cooperate), it will be readily apparent that a single device/article may be used in place of the more than one device or article or a different number of devices/articles may be used instead of the shown number of devices or programs. The functionality and/or the features of a device may be alternatively embodied by one or more other devices which are not explicitly described as having such functionality/features. Thus, other embodiments of the present invention need not include the device itself.

The foregoing description of various embodiments of the invention has been presented for the purposes of illustration and description. It is not intended to be exhaustive or to limit the invention to the precise form disclosed. Many modifications and variations are possible in light of the above teaching. It is intended that the scope of the invention be limited not by this detailed description, but rather by the claims appended hereto. The above specification, examples and data provide a complete description of the manufacture and use of the composition of the invention. Since many embodiments of the invention can be made without departing from the spirit and scope of the invention, the invention resides in the claims herein after appended.